WO2018050802A1 - Method for classifying spectra of objects having complex information content - Google Patents

Abstract

The invention relates to a method for classifying spectra of objects having complex information content after registration of the spectra by using a method for pre-processing data and by using a method, associated with the data pre-processing, for classification with the calculation of a classifier. After the registration of the spectra and the pre-processing of the spectra, a multiple classification method is performed, having at least two different methods of the data pre-processing of the spectra and of the method, associated with the particular data-preprocessing, for classification. After the registration and the data pre-processing of the spectra, the following steps are performed: calculating a plurality of classifiers of the series per type of data pre-processing, determining the classifiers of the series with iterative adjustment and validation, calculating probabilities of the class association, all classifiers of the series or classifiers being equally incorporated in the determination of a classification result.

Description

 Method for classifying spectra of objects with complex information content

The invention relates to a method for classifying spectra of objects having a complex information content with at least two different object information, in particular of optical molecular spectra for the assignment of the object information.

There are known publications on methods for the supported classification of optical spectra. These include u.a. The document A. E. Nikulin, B. Dolenko, T., Bezabeh, R.L. Somorjai: Near-optimal region selection for feature space reduction: novel preprocessing methods for classifying MR spectra, NMR Biomed. 1 1 (4-5), 1998, pp. 209-216, BK Lavine, CE Davidson, AJ Moore's: Genetic Algorithms for Spectral Pattern Recognition, Vibrational Spectroscopy, Volume 28, Issue 1, 2002, Pages 83-95, in which the algorithm is based on the main components and a classification of spectral ranges is used for the classification, and the publication J. Jacques, C. Bouveyron, S. Girard, O. Devos, L. Duponchel, C. Ruckebusch: Gaussian mixture models for the Classification of high-dimensional vibrational spectroscopy data, Journal of Chemometrics, Volume 24, Issue 1 1 -12, pp. 719-727.

It describes a method in which particularly high-dimensional spectral data are split into so-called subspaces, which are subsequently classified by means of discriminant analysis. Optical molecular spectra have a high information content about the molecular properties of the examined object. In particular, vibration spectra are considered to be the fingerprint of molecules because of their high information density over the molecular structure. In the spectroscopic analysis of complex biological objects, the information relevant to the specification must be separated from the less or insignificant information as well as from interferences. These are usually procedures Chemometrics, also multivariate methods used in higher-dimensional data.

 If important spectral features of the desired molecular information of objects are known, supported (supervised) classification methods can be used, as for example in the publication G. Steiner, S. Kuchler, A. Herrmann, E. Koch, R. Salzer, G. Schackert, M. Kirsch: Cytometry, Part A 2008, 73A, 1 158-1 164.

The supported classification methods are distinguished from other methods by a higher accuracy in the detection and quantitative evaluation of the information sought. In the known method of the supported classification according to FIG. 4 a, a classifier 50 is calculated by means of a training set 19, which consists of representative spectra with assigned properties. The classifier 50 is then evaluated on the basis of an independent test set 29, i. the spectra are not used to construct the classifier 50, as shown in FIG. 4 a, validated and evaluated for validation, and achieved a classified test set 24.

 The construction of the classifier 50 by means of the training set 19 is performed by a test set 29 with e.g. a maximum of 30% of the spectra (dashed line to the created classifier) is verified according to FIG. 4 b, so that, as a result, a classified test set 24 (dashed line from the classifier 50) to the classified test set 24 is achieved.

A general problem of the supported classification methods is the trade-off between the accuracy of the assignments obtained and the robustness of the classification. Often very high accuracies can only be achieved with a so-called overtraining of the classifier. This means that the classifier can only correctly assign certain spectra, with very high accuracy. By contrast, even the slightest deviation or interference leads to a dramatically reduced accuracy of the Classification. An attempt is therefore made to adapt between best possible classification and high robustness of the classification.

If spectra with very high variability are present, as for example in in-ovo species for determining the sex of chicken eggs, it is inevitable that accuracy will be compromised to ensure sufficient robustness of the classification. Basically, this immanent contradiction can not be solved. Nevertheless, in order to achieve good stability with sufficient accuracy, various methods of classification have been newly developed in recent years. The basic approach is the parallelization of the classification across different decision trees. The Random Forest method is based on a network of uncorrelated decision trees, where the decision trees are grown by randomization during the training process. Every so-called tree makes a decision. The group of trees with the most identical decisions determines the result of the classification, ie the assignment of the spectrum. However, the Random Forest method can not respond to different disturbances or variations in the spectra. Again, by putting up too many trees, overtraining can occur.

The document US20120321 174 A1 describes a classification method based on the Random Forest method for image evaluation. This supported classification method is designed to take into account, in particular, small but relevant features for classification.

These relevant features of the general classification method can also be defined and play a role, for example, in the in-ovo spectroscopy of chicken eggs in the form of small spectral-related signals of the sex information. In in ovo spectroscopy of chicken eggs a supported classification method is used to identify the sex.

However, optical in-ovo spectra are often characterized by a very high natural variability, which clearly overlap the comparatively small signals of gender information. In addition, there are unavoidable external influences from the environment of the measurement itself.

At present, the following different methods for classifying the spectra of objects, in particular for the sex determination of fertilized and / or incubated eggs, are given in the documents cited below:

The document WO 2010/150265 A describes a method based on a coloring, in particular of the feathers of the developed embryo. The method is based on the fact that in the advanced stage of development (day 12 of incubation), the color of the feathers in certain breeds of chicken allows conclusions about the sex. The evaluation is carried out with an algorithm for classification. Furthermore, the document WO 2014/021715 A2 describes a method in which the sex of the embryo is determined by means of endocrinological analysis.

The document DE 10 2007 013 107 A1 describes the application of Raman spectroscopy for sexing of birds, wherein generally cell-containing material is examined. However, no method for in ovo sex determination is described.

The molecular spectra are registered by means of methods and devices according to the following publications: A method and apparatus for determining the sex of chicken eggs, based on the optical, preferably fiber-coupled spectroscopy are described in the document DE 10 2010 006 161 B3. However, no methods for spectral analysis and classification are described

The documents DE 10 2014 010 150 A1 and WO 2016/000678 A1 describe methods and devices for Raman spectroscopy in ovo sex determination. The evaluation of the spectra can advantageously be carried out using chemometric methods.

Document EP 2 336 751 A1 describes a method for determining the sex of bird eggs. In the method, the germinal disc of an egg is illuminated with light and the emitted fluorescence detected time-resolved. The recognition of the gender is done by means of a supported classification, whereby a classifier is calculated by means of the method of the fractal dimension.

The document US 6 029 080 B describes a method for in ovo sex determination. From the analysis of MRI images of the egg, the sexual organs can be identified and used for sex determination at a certain developmental stage of the embryo.

The disadvantages in the evaluation of these methods is that each of these methods ultimately uses a method for classifying spectra of objects with only one classifier.

To summarize, the drawbacks are that in order to safely extract the searched gender information from the registered spectra, therefore, the use of only a single classifier is insufficient to consider the recognition certainty of the particular sex information sufficient. Rather, the variable influences and the variations in the biochemical composition of the egg as well as in the different stages of development. In order to incorporate this large range of variations in a recognition-proof method for classification, it is estimated that therefore the calculation of only one classifier will not be sufficient.

The invention is therefore based on the object of specifying a method for the classification of spectra of objects with complex information content, which is designed so that a maximum accuracy of the determination of the associated selected features of objects is achieved, at the same time at least maintain the stability of the classification should stay. Thus, an adaptation between best possible classification and high robustness of the classification should be sought. At the same time an overtraining of the classifier should be avoided.

The object is achieved with the features of claim 1.

In the method for classifying spectra of complex information content objects with at least two different object information, using a method for registering and pretreating spectral data and a data pretreatment associated method of classification with the calculation of a classifier, according to the characterizing part of claim 1 after the registration of the spectra and the pretreatment of spectral data, a multiple classification method is carried out with at least two different methods of data pretreatment of spectral data and a classification method associated with the respective data pretreatment.

Within the scope of the invention, the recording, determination and storage of spectra and the generation of spectra are intended to include the registration of spectra digitized signals are available for storage, which are available for the further data pretreatment of the spectral data.

Depending on the pre-treatment algorithm used, the data pre-treatments generate different corrected pretreated spectra with many data points, which are assigned to at least one classification procedure.

The following steps are performed after registration and data pre-treatment in the evaluation procedure of registered spectra of objects:

a calculation of several classifiers of series per type of data pre-treatment,

 a determination of the classifiers of the series, iteratively calculated and validated,

- a calculation of probabilities of class membership,

- an equal inclusion of all classifiers of the series or

 Classifiers in the determination of a classification result.

In determining and determining the number of calculated classifiers N _G in the series per classification group, both the circumference of the spectral data points vs and the double half width of the spectral regions ws and the number of selected spectral regions R _s for the classification in equation (I ) considered:

the equation (I) ensures that each data point v can be selected with equal probability.

However, the data points belonging to a circumference of the spectral data points can also be weighted. At least one of the spectral pretreatments is structured in such a way that particular features emerge and other specific features are suppressed, so that different characteristics are used for the classification.

At least one pretreatment of spectra can be designed with the same specific features and at least one of the abovementioned spectral pretreatments with differently determined characteristics can be used for the classification. The pretreated spectra can be interpreted as variable training sets, and several classifiers of the series or classifiers are determined and validated iteratively.

In the context of the invention, the classification is to be understood as the classification of the pretreated spectra into a respective class determined according to a predetermined algorithm. The method of classification is carried out with the aid of predetermined parameters and the result of the classification is expressed by a calculated classifier. At least one supported classification and / or non-assisted classification method may be used to select spectral regions or single wavelength ranges and subsequent analysis. A linear or nonlinear discriminant analysis can be used. For classification, methods of neural networks and / or a method of linear time-frequency transformations (English: Wavelets) can be used.

The spectra of optical molecular spectroscopy, such as absorption, emission, scattering or UV / vis, NIR, IR absorption, fluorescence or Raman, can be classified here. As data pre-treatments of the registered spectral data or raw spectra, baseline corrections, normalizations, derivatives, covariance and / or a principal component analysis can be used.

To evaluate the classifiers of the series for a classification result, a calculation of a median or a performance of a cluster analysis can be provided.

 The median or central value is given as an average for distributions in the statistics. The median of a list of numeric values is the value that is at the middle (central) location when the values are sorted by size. The value of the size here represents in each case the point value of a classifier or the probability of the class membership determined by the classifier.

The method according to the invention can generally be completed in detail with the following steps:

 Recording and registration of the spectra by means of at least one optical device with at least one spectrometer and / or further detectors,

 Generation of digitized signals in the form of data points and storage of the registered spectra in storage units of classification units of an evaluation unit,

 Spectral pretreatment by individually pretreating the registered and stored spectra in the individual storage units and providing the associated digitized evaluated signals for further processing,

 Separation of the pretreated spectra as a training set and as a test set,

- design and use of the pretreated spectra as a training set and of a test set separate from the training set,

wherein according to the invention at least one - Calculation of the classifiers of the series of individual classification methods considered, including iterative procedures and validation in the classification groups,

 Classification of the pretreated spectra of the training set with all classifiers of the series,

 Classifying the spectra of the training set into a class of object information with an expression of a class membership probability,

 Calculation of a classification result by calculation of the median or by carrying out a cluster analysis to represent the

Likelihood result of the object information belonging to the class of the training set,

 Classification of the pretreated spectra of the test set with all classifiers of the series,

 Classifying the spectra of the test set into a class of object information with an expression of a class membership probability; and

 Calculation of the classification result by calculation of the median or by performing a cluster analysis to represent the probability result of a class

Object information of the test set

 be performed.

As objects with object information any bird eggs, optionally chicken eggs can be used and as object information can be used in a specific application, the dual information about the female or the male sex.

The method thus includes steps for performing a multiple classification based on different conventional evaluation methods after a spectral detection and registration downstream Spectral pretreatment and a subsequent multiple calculation of different classifiers. In this case, at least one spectra pretreatment is involved, in which the spectral pretreatment is structured in such a way that, taking into account the equivalence of features, consideration of particular features is more pronounced and other features are more strongly suppressed. The pretreated spectra are now designed as a training set, whereby several series of classifiers are calculated. The classifiers are calculated and validated iteratively. In this way several classifiers can be determined. The spectra of the test set are subsequently classified with all classifiers. The classification of the spectra into a specific class of object information / features (in chickens: male, female) is preferably done as a score or an expression of a class membership probability. In order to obtain a sole statement from the classifiers, the relationships within the classifiers are determined. A simple means for ratio representation is this, for example, a calculation of the median or a cluster analysis.

An apparatus for classifying spectra of objects having complex information content, preferably with objects in the form of chicken eggs for a determination of dual egg information - female or male -, wherein the device performs the aforesaid method,

may include at least the following units

 at least one detecting optical device having at least one spectrometer and / or further detectors for recording and registering the spectra,

 - A unit for generating digitized signals in the form of data points that realize the spectra.

Storage units for storing the registered spectra in the classification units / groups of an evaluation unit comprising the classification units, - units for spectral pretreatment, in which the registered spectra are individually pretreated in the individual storage units and the associated digitized evaluated signals - the data points determined - are made available for further processing,

 - Training sets for the interpretation and use of pretreated spectra,

 At least one classification unit for the groups for the calculation of the classifiers of the series of the considered individual conventional classification methods, including iterative methods and a validation in the classification units,

 - test sets for the classification of the pretreated spectra with all classifiers of the series,

 a unit for classifying the pretreated spectra into at least one dual class - in chickens: male or female - of object (egg) information with an expression of probability

Class membership,

 a rating unit for calculating the classification result e.g. in the form of the median or by performing a cluster analysis to determine the probability of at least one of the dual class - in chickens: female or male - associated

Object (egg) information.

Further developments and further embodiments of the invention are specified in further subclaims.

The invention will be explained by means of exemplary embodiments with reference to drawings.

Show it:

Fig. 1 is a schematic block diagram of a method according to the invention for the classification of spectra of objects with complex Information content, in particular of optical molecular spectra for the assignment and determination of dual object information, the method being implemented as a multiple classification method, FIG. 2a a schematic representation of a single classification procedure with spectral pretreatment: raw spectra, taking all features into consideration as equal (circles of equal size),

2b shows a schematic representation of a single classification method with spectral pretreatment: linear baseline correction, with a favored large feature circle of the fluorescence intensity and several less relevant small feature circles,

Fig. 2c is a schematic representation of a single classification method with spectra pretreatment: normalization, with a favored large

Feature circle of the fluorescence profile and several less relevant small feature circles,

2d is a schematic representation of a single spectrally pretreatment classification method: Raman spectra, with a favored large feature set of molecular composition and several less relevant small feature circles.

Fig. 3a is a schematic representation of the Rohspektrum classification method associated spectra of FIG. 2a, wherein the punctured

Spectrum is assigned to the female chicken egg spectrum,

3b shows a schematic representation of the spectra associated with the linear baseline correction classification method according to FIG. 2b, the dotted spectrum being associated with the female chicken egg spectrum; 3c shows a schematic representation of the spectra associated with the nomination classification method according to FIG. 2c, wherein the dotted spectrum is assigned to the female chicken egg spectrum,

FIG. 3d shows a schematic representation of the spectra associated with the Raman spectrum classification method according to FIG. 2d, wherein the dotted spectrum is assigned to the female hen's egg spectrum, FIG. 4a a schematic representation of the sequence of a classification with FIG

 Determination of a classifier according to the prior art,

4b shows a flowchart for the multiple classification method according to the invention with training set and test set in algorithmic connection with classification result design from a large number of classifiers,

FIG. 5 is a schematic representation of a probability / classifier column representation for twenty classifiers of FIG. 1 for a chicken egg used as an example, with the one above the dashed line. FIG

Line - separating boundary - lying columns of the female

 Be assigned to gender,

6 is a plan view of the pillar representation of an egg according to FIG. 5 and a possible view on a display,

FIG. 7 shows a probability (female) / classifier number representation with the indication of the calculated median for an egg with the 20 classifiers according to FIG. 6, wherein a bold-dashed separation limit at 0.5 .mu.m

 Probability and a thin-dashed median at about 0.72 of the

Probability "female" lie, so the sex of the egg can be identified as female,

Figure 8 is a schematic representation of a probability / classifier number column representation for an egg for an optional 120 classifiers, where the end portions of the columns above the bold dashed line boundary are the female gender of an egg and the end portions of the columns below the fat dashed separation limit FIG. 9 shows a representation of the calculated median, shown dashed, for an egg recognized as female with the 120 classifiers according to FIG. 8 in a probability / classifier number representation, FIG.

Fig. 10a is a schematic representation of a single

Spectral Pretreatment Classification Procedure: Raw spectra, taking into account all features as equal, with eight selected spectral regions Rsi to Rse at a spectral data point per area spectral range between wavenumbers 570 cm ^"1 to 2750 cm ^{" 1} to determine the number of classifiers per data pretreatment,

10b shows an enlarged detail (l-1) of the data point representation in FIG

give spectral region R _s 8 according to FIG. 10 a, FIG. 10 c shows an enlarged detail (11) of the data point representation in FIG

 spectral region Rse with an indication of the weighting of data points in the region of the region Rse according to FIGS. 10a and 10b,

1 a is a representation of the calculated median, shown dashed, for an egg recognized as male with the 120 classifiers similar to the one in a probability / classifier number representation, FIG. 11b shows a first histogram representation of the dependence between the number of elements of the cluster and the center of gravity of the cluster; and FIG. 11c shows a second histogram representation of the dependence between the FIG

 Number of elements of the cluster and the center of gravity of the cluster.

In the following, Fig. 1 and Figs. 2a, 2b, 2c, 2d are considered together. 1 is a schematic block diagram of a method 1 according to the invention for classifying spectra 4 of an object 2 having a complex information content with at least two / dual and different object information / features, in particular optical molecular spectra 4 for assigning object information / features 3 a probable determination of example dual object information 31, 32 is shown.

Bird eggs, for example chicken eggs, can be used as objects 2 to be examined, and as dual object information 3, for example, the feature 31 can be searched and determined via the female gender and the characteristic 32 via the male gender.

The following is the description of the method 1 according to the invention for carrying out the classification.

 For this purpose, a block-wise sequence of the method 1 according to the invention is shown in FIG.

In the method 1 for classifying spectra 4 of objects 2 having complex information content with at least two different object information, after the registration using a method of pretreating data and a method of classification associated with the data pretreatment, the calculation of a classifier takes place. According to the invention, after registration and data pretreatment of spectra 4, a multiple classification method with at least two different methods of data pretreatment 5, 6, 7, 8 of the spectra 4 and the method for classification in groups 9 assigned to the respective data pretreatment 5, 6, 7, 8 10, 1 1, 12 for determining a plurality, eg five classifiers per group 9, 10, 1 1, 12, that is from a total of many, eg twenty (five classifier / group x four groups) classifiers 131, 132, 133, 134, 135, etc. performed for the series 14, 15, 16. The following steps are carried out after registration and data pretreatment of the spectra, the steps relating to FIG. 1:

 a calculation of five classifiers of series 13, 14, 15, 16 per type of data pre-treatment 5, 6, 7, 8 so that finally twenty classifiers 131, 132, 133, 134, 135, etc. are determined,

- a determination of the five classifiers of series 13, 14, 15, 16, iteratively adapted and validated,

 - a calculation of probabilities of class membership,

 an equal inclusion of all five classifiers of the series 13, 14, 15, 16 or classifiers 131, 132, 133, 134, 135 etc. into the determination of a classification result 18, e.g. in the form of a median 30.

In determining or determining the number of classifiers N _G to be calculated in the series 13, 14, 15, 16 with respect to the groups 9, 10, 11, 12, a circumference of the spectral data points vs and a double half-width w _s of spectral regions R _s as well as a number of the selected spectral regions R _s in the following equation (I):

N _c = -

2w _s - R _s ^, where equation (I) ensures that each data point vs can be selected with equal probability. For a total of twenty classifiers of the four series 13, 14, 15, 16 with N _G (13), Q (14), N _G (15) and N _G (16) according to FIG. 1 as well as according to FIG. 5, FIG. 6 and FIG. 7, the following parameters are given for the entire spectral range of 500 cm ^"1 to 2750 cm ^{" 1,} for example:

Scope of the spectral data points vs in a given entire spectral range 500 cm ^"1 to 2750 cm ^{" 1} with v _s = 800,

Number of selected spectral regions R _s with R _s = 8,

Width B of the spectral regions R _s with B = 2 ^■ w _s = 5, ie in a region R _{s there} may be twenty data points vs. The half-width ws is thus ws = 2.5.

The spectral pretreatments 6, 7, 8 can be structured in accordance with FIGS. 2b, 2c, 2d such that particular features are favored and stand out and other features are suppressed.

In the spectral pretreatment 5 of the raw spectra 25 according to FIG. 2a, all the features taken into consideration can be treated as equal and pretreated. The pretreated spectra 4 are designed as a training set 24 according to FIG. 4b and several classifiers of the series 13, 14, 15, 16 or eg in detail for series 13 the classifiers 131, 132, 133, 134, 135, etc. iteratively determined and validated , After passing through at least two classifying processes 9, 10, 11, 12, each with a preceding spectrum-different data pretreatment 5, 6, 7, 8 with at least one determined classifier 131, 141, 151, 161, at least two of the determined classifiers can be obtained as a whole 131, 141, 151, 161 for the evaluation and the subsequent determination of a probability result 18 with respect to the predetermined different object information 31, 32 achieved and are used, wherein the probability result 18 is output, so that a conclusion is made possible at least on the highest value determined object information 31 or 32. At least one method of supported classification and / or unsupported classification may be used to select spectral regions R _s or single wavelength ranges / wavenumber ranges and subsequent analysis.

 The subsequent analysis may be a linear discriminant analysis or a non-linear discriminant analysis.

However, it is also possible to use as a method for classification in groups 9, 10, 11, 12 a method of neural networks and / or a method of linear time-frequency transformations (English: Wavelets).

The spectra 4 of the optical molecular spectroscopy, such as absorption, emission, scattering or UV / vis, NIR, IR absorption, fluorescence, Raman can be classified by the method according to the invention. For the data pretreatments 5, 6, 7, 8 shown in FIG. 1, raw spectra 25, baseline corrections 26, normalizations 27, derivatives, covariance and / or a principal component analysis / Raman spectra 28 can be defined and used. A data pretreatment consists in the formation of digital signals, that is to say of data points which, when lined up, yield the respective calculated spectral curve of 25, 26, 27, 28 and which can thus be assigned to the individual different classification methods used.

For evaluating the classifiers of the series 13, 14, 15, 16 for a classification result 18, a calculation of a median 30 (FIG. 7, 9, 11 a) for an object 2 or a feedthrough (FIG a, Fig. 1 1 b) are provided a cluster analysis. As an example of an evaluation, the known k-means cluster analysis can be used. In this case, at least two clusters for "male" or "female" are given in FIG. 11a. The cluster to which most of the elements, ie probabilities, are assigned, defines the gender (Fig. 11b-male). A cluster is a group of composite elements with similar properties (here: probabilities - classification results). 1 a shows a curve of the dependence between probability with a characteristic "male" and a classifier number of 120 classifiers The plot of the sorted probabilities shows that the median 30 is just above the separation limit 42 of 0.5 of the probability coordinate Thus, the egg 2 is barely classified as male, and the well-known k-means cluster analysis leads to a clearer result.

1 1 b, a first histogram representation 43 of the dependence between the number of elements of the cluster and the center of gravity of the cluster, and FIG. 11 c a second histogram representation 44 of the dependence between the number of elements of the cluster and the center of gravity of the cluster, whereby the achieved classification results apply as elements.

 For this purpose, 1 b two clusters are formed in Fig. 1, the centers of gravity are 0.84 and 0.17. Because the cluster with the centroid 0.84 has more elements, i. Classification results are assigned, the egg 2 can be rated as clearly male.

In the selection of five clusters in the histogram representation 44 according to FIG. 11c, the result is also confirmed. Here 65 calculated probability values are classified as male, whereby the cluster (number 4 = # 4) with the strongest value 0.96 for "male" also contains most of the elements according to Fig. 11a, so that the gender of the egg 2 is clearly "male".

The same and the same applies to the cluster analysis, if the sex of the egg 2 is determined as female.

 The method 1 according to the invention can be realized by means of the following steps using hardware components of an associated device:

 - Recording and registration of the spectra 4 by means of at least one optical device with at least one spectrometer and / or further

detectors,

 Generation of digitized signals, the data points, and storage of the detected spectra 4 in storage units of the classification units of an evaluation unit,

 Spectral pretreatment 5, 6, 7, 8, in that the stored spectra 4 consisting of the data points vs are evaluated individually in the individual storage units and the associated digitized evaluated signals are made available for further processing,

- design or formation of the pretreated spectra 25, 26, 27, 28 as a training set 19 and as a separate test set 29,

 - Calculation of classifiers of series 13, 14, 15, 16 in the form of individual classifiers 131, 132, 133, 134, 135 of a series 13, etc. of the considered individual classification methods 9, 10, 1 1, 12 involving iterative methods and a validation in the classification units / groups,

 Classification of the evaluated spectra 25, 26, 27, 28 of the test set 24 with all classifiers of the series 13, 14, 15, 16,

Classifying the spectra 25, 26, 27, 28 into a class of object information with an expression of a class membership probability, Calculation of the median 30 or execution of a previously mentioned cluster analysis to display the probability result in the form of a classification result 18 of the object information associated with a class.

The construction of classifiers with respect to the series 13, 14, 15, 16 by means of a training set 19 is performed by a test set 29 with e.g. verified as a maximum of 30% of the spectra (dashed line to the classifiers 13, 14, 15, 16) according to FIG. 4b, so that as a result a classified test set 24 (broken line from the classifiers of the series 13, 14, 15, 16 to the classified Test set 24) is achieved.

It should be noted that in general the registered in-ovo spectra 4 are naturally highly variable. This is due, on the one hand, to the inherent variability of biological systems and, on the other hand, to the sensitivity of Raman spectroscopic measurements.

 External perturbations of a systematic and random nature lead to a high variability of the spectral features and thus superimpose the gender-relevant information.

Furthermore, in the method of Raman spectroscopy also fluorescent light occurs, which also contains molecular information, but at the same time superimposes the generally much weaker Raman spectroscopic molecular information on the composition of the examined object 2. FIG. 2 a shows a schematic representation of the raw spectra 25 as one of all considered individual classification methods in FIG. 1 of the four individual classification methods.

In general, according to FIGS. 2 a, 2b, 2c and 2d, at least four classes of signals or specific features can be formed:

molecular composition 20, Fluorescence intensity 21,

 - Fluorescence profile 22 and

 - Variation of physical parameters 23, wherein these specific features 20, 21, 22, 23 are formed for visual representation in Fig. 3a, Fig. 3b, Fig. 3c and Fig. 3d as the same large bordered and / or different dashed lines surrounded circles.

Behind each classifier is a mathematical expression for the separation of the signals according to the object information 3 (31 female, 32 male).

Three classes / features 20, 21, 22 of the four classes / features 20, 21, 22, 23 contain gender-relevant information. However, it is not possible to eliminate the variation 23 of the physical parameters from the spectra 4 such that no or only a small loss of information arises in the three other classes 20, 21, 22. The raw spectra 25 thus have the highest content of all information, but also the highest content of disturbances due to the equality of all certain characteristics. By adding at least one of said data pre-treatment e.g. 26 from the data pre-treatment 26, 27, 28 with differently valued features, the interference is reduced. By using additional data pre-treatments 27, 28, the original disturbances are minimized or even eliminated.

It is shown in FIG. 2b that the in-ovo spectra 4 registered as digital signals are subjected to a linear baseline correction 26, whereby the fluorescence intensity signal 21 (large circle) stands out. At the same time, signals 23 of physical parameters in the spectra are suppressed (small circle). Due to the usually large differences in intensity between fluorescence signal and Raman signal, the information on the molecular Composition 20 (small circle) in the background. However, the fluorescence intensity 21 (large circle) itself is a potential marker for sex recognition because male embryos often, but not always, have a biochemical composition in the blood that has increased fluorescence intensity 21 over female embryos or female embryo blood.

FIG. 2c shows that by means of the methods of spectral normalization 27, for example by means of vector or surface normalization, it is possible to compensate for variations in the fluorescence intensity 21 (small circle) and to minimize the random influences of physical parameters 23 (small circle). Thus, the fluorescence profile 22 (large circle) can preferably be emphasized. At the same time, only a small amount of molecular composition information 20 (small circle) based on Raman signals is minimized. Since the fluorescence profile 22, ie the spectral characteristic of the fluorescence, is determined by the molecular composition 20, information relevant to gender can be highlighted.

FIG. 2d shows that as complete a correction as possible of the so-called background of Raman spectra 28 leads to the sole highlighting of the Raman bands, that is to say the information about the molecular structure and composition 20 (large circle) of the examined object 2.

FIGS. 3a, 3b, 3c and 3d each show a schematic representation of the spectra associated with the individual classification methods (relative to the relative wavenumber) with reference to FIGS. 2a, 2b, 2c, 2d.

In this case, at least the spectral pretreatment 5 with identically determined features of at least one of the spectral pretreatments 6, 7, 8 with differently determined features is added to the evaluation. 4b shows a flow chart for the multiple classification method according to the invention with training set and test set in local separation, but in algorithmic connection with a classification result design from a large number of classifiers. In this case, the respective spectral pretreatment is structured so that in each case certain features stand out more clearly and other features are more strongly suppressed. According to FIG. 4b, the pretreated spectra 4 are now designed as a variable training set 19, with several series 13, 14, 15, 16 of classifiers being calculated, for example, in detail of a series 131, 132, 133, 134, 135 and so on. Usually all classifiers of series 13, 14, 15, 16 are calculated and validated iteratively. Variability means that without further preconditions any spectra can be selected for each classifier to be calculated. In this way, for example, several classifiers 131, 132, 133, 134, 135 for the first series 13, etc. can be determined. This also applies to the other series 14, 15, 16. The selected spectra of the test set 29, eg 30%, are classified according to FIG. 4b below with all classifiers to a classified test set 24. Classification of spectra 4; 25, 26, 27, 28 into a certain class of the characteristics (male, female) is preferably carried out as a score or in an expression of a probability for class membership. In order to obtain a sole statement from the classifiers of the series 13, 14, 15, 16 or 131, 132, 133, 134, 135, etc., the relationships within the classifiers 13, 14, 15, 16; 131, 132, 133, 134, 135 determined. A simple means of doing this is to compute the median 30 or, as stated previously, perform a cluster analysis.

In the flow chart shown in Fig. 4b, at the point indicated by 45 = ein, a comparison is made of each classified spectrum of each form of pretreatment with the feature. As a result, only "correct" or "wrong" is output.

Example: The training set 19 comprises 100 spectra. Of these, 60 are selected for the calculation of the classifiers. If four methods of data pre-treatment 5, 6, 7, 8 are used, there are 60 x 4 = 240 classified spectra. From the comparison with the feature list, 240 statements result either "right" or "wrong". This result is achieved, for example, in the specified 129th iteration stage.

In the point denoted by the reference numeral 46 = ©, an evaluation of the classified spectra is carried out with respect to a specified criterion or several specified criteria. The criteria used are, for example, a limit of accuracy or a maximum number of iterative steps. The criteria may be logical AND or logical OR linked.

 Example: Of the possible 240 statements, 205 are "right" and 35 are "wrong". This results in a correctness for the training set of 85%.

Before the start of the classification, criteria are set as:

 1st accuracy> 80% and

 2. a maximum number of iterations: 1000.

that is, after the finished

 129. iteration stage <1000

 and at

an accuracy of 85% achieved> given correctness.

If logical AND can be "bad" (but the classifiers as the best

Intermediate result are saved) and

with logical OR, "good" can be achieved.

If the number of predefined classifiers to be determined is reached at the point indicated by reference symbol 47 = ©, all classifiers (each having led to the best result at point © = 45) are passed to the validation of the entire training set 19 , Example: It is prescribed to calculate 30 classifiers per data pre-treatment 5, 6, 7, 8 and lead to a multiple classification. This will pass 30 x 4 = 120 classifiers for validation. At the point / reference junction indicated by the reference symbol 48 = ®, a joint evaluation of the classification of all spectra of the training set 19 is carried out by the method of leave-one-out or cross-validation.

In the case of an "passed test", the classifiers are passed to classify the "unknown" spectra of the test set 29.

If the test fails, classification is not possible according to the given criteria.

In the point / reference junction indicated by the reference symbol 49 = ©, a final evaluation of the classification of the spectra of the test set 24 is carried out on the basis of its known features.

Example:

 The test set 29 comprises 50 spectra. These were each classified with 120 classifiers, i. Each spectrum is assigned 120 probabilities for class membership. According to the median or the cluster analysis, this results in belonging to a class. This is the result of the multiple classification for each individual spectrum. For example, if 41 of the 50 spectra are correctly classified, this results in an accuracy of 82% for the entire test set 24. From the comparison with the feature list, the thus created method 1 of the multiple classification is finally evaluated. Thus, the method is created and can now be used for spectra without knowledge of the features.

FIG. 5 shows a schematic representation of a probability / classifier number column representation 38 for 20 classifiers according to FIG. 1 for the display of an egg 2 identified as female. This can be the unshaded End portions / end faces 33 of columns 34 belonging to the female gender above a certain line - the separation boundary 42 - lie and the columns 35 lie with the hatched end portions / end faces 36 as the male sex belonging to the separation boundary 42. The separation limit value is 0.5 in FIG. 5 and the median 30 has the value 0.72 there. Thus, the egg 2 is classified as clearly "female".

Shown in Fig. 6 is the perspective view associated with the perspective view of the column, which is indicated on a color display as the classification result image 37 with the plurality of unshaded end areas / end faces 33 for the object information 31 "female." On the color display, the unshaded faces 33 red and the hatched end faces 36 may be blue, so that a color visual representation of the rating of the sex can be made.

The shaded squares can be displayed in blue and the unshaded squares in red. The few blue squares indicate the male object information 32. The predominantly red squares indicate the female object information 31. Since the red squares predominate, the gender of the incubated chicken egg 2 can be identified as a female trait 31.

FIG. 7 shows a representation of the calculated median 30 for 10 classifiers in relation to the number of twenty classifiers for an egg 2 according to FIG. 1 with a series of five classifiers 131, 132, 133, 134, 135 per group in four groups 9, 10, 1 1, 12 indicated. In the column and median representation, 17 classifiers give the indication of a female egg 2. The total classification result 18 can be given with the calculated median 30. FIG. 8 is a schematic representation of another exemplary probability / classifier number column representation 39 for 120 Classifiers, shown as columns, for displaying an egg 2 identified as female. The separation boundary 42 also shows the boundary between the feature "male" and the feature "female". Here, too, the columns 34 (31) ending above the separation boundary 42 are hatched on the end faces and the hatches 35 (32) ending below the separation boundary 42 are indicated by hatching.

In an egg 2 with male gender feature 32, another column representation may be formed, in which case the above the separation boundary 42 lying end faces of the formed columns 35 in their majority compared to the non-hatched end faces of the columns 34 are hatched (not shown).

FIG. 9 shows a plot of the calculated median 30 in relation to the number of total 120 classifiers according to the column representation in FIG. 8 in a probability / classifier representation for a female gender feature 31 with classifiers sorted by increasing points. Here, the median 30 is half of the calculated 120 classifiers and has a probability of 0.95.

The classification units / groups 9, 10, 11, 12 for determining the object information in the form of dual sex characteristics 31, 32 - female or male - of fertilized and non-hatched and incubated eggs 2, which are contained in an evaluation unit, function as follows:

It explains the operation.

From each class 25, 26, 27, 28, several classifiers of the series 13, 14, 15, 16 are calculated after the spectral pre-treatment 5, 6, 7, 8. The determination of the classifier series 13, 14, 15, 16 takes place according to an algorithm which initially selects spectral regions R _s from the coordinate of the relative wavenumbers in a kind of tandem method and subsequently classifies the intensity values of the selected regions R _s by means of discriminant analysis. In comparison with the training data of the class affiliation, a selection of spectral classes and the classification of the intensity values takes place in a repeated step again. This cycle is repeated iteratively until reaching a no longer improvable accuracy, wherein the termination criterion can be specified.

 The risk of over-training and thus the achievement of high instabilities increases the more spectral classes 25, 26, 27, 28 are used for the classification. Therefore, it is desirable to use only a few (3 to a maximum of 20) spectral classes for the creation of the classifier series 13, 14, 15, 16. However, since the gender-relevant information is distributed over the entire spectral range, albeit differently, even the creation of just one classifier would even make essential spectral information unused. For this reason, it is expedient that several (10 to 30) classifiers are calculated in the series 13, 14, 15, 16 per group of data pretreatment 5, 6, 7, 8.

This has the advantage that on the one hand the accuracy of the classification is improved, based solely on the fact that as much spectral information as possible is included and on the other hand the robustness, i. the stability is increased, since several classifiers of the series 13, 14, 15, 16 carry the assignment and individual misalignments are compensated.

The hardware units assigned to the classifications work the same for all four groups 9, 10, 11, 12. Thus, instead of the four units driven in parallel, only one unit can be used which serially creates the series 13, 14, 15, 16 of the classifiers in a predetermined sequence. In determining the number of calculated classifiers N _G in the series 13, 14, 15, 16 per group 9, 10, 11, 12, the circumference of the spectral data points v _s and the double half width of the spectral regions ws and the number of the selected spectral regions R _s :

Equation (I) ensures that each data point vs can be selected with equal probability.

In FIGS. 10a and 10b, twenty classifiers per data pretreatment 25 are given using the example of the raw spectra (intensity / wavenumber curves). For this purpose, a section of the associated male intensity / wavenumber curves is shown enlarged in FIG. 10a. The circumference of the spectral data points v _s in the entire spectral range between 500 cm ^"1 and 2750 cm ^{" is 1} data point. The number of selected spectral regions R _s in FIGS. 10 a and 1 b is Rs = 8 with R _s i, R _s 2, R _s 3, R _s 4, R _{s s} , R _s 6, R s 7, and R s

From this, twenty classifiers N _G for the raw spectrum 25 can be calculated according to equation (I). This means that in four data pre-treatments 25, 26, 27, 28 a total of 80 generated (20 classifiers / group x 4 groups) classifiers.

After the enlarged section in FIG. 10 c, the data points v _{s can} also be weighted additionally. For this purpose, a weighting diagram 40 is indicated from which it can be seen that the highest weighting value is assigned to the middle data point 41.

 This can be done with both the male spectrum and the female spectrum.

The rating 17 and the classification of the results assigned to the classifiers of the series 13, 14, 15, 16 are performed in a rating unit and led to a classification result 18 (30). Finally, a classification result 18 is output in the form of the median 30, which represents the sexually most probable dual sex information 31, 32 (male or female) in the sexing of chicken eggs.

The method according to the invention can generally be completed in detail with the following steps:

 Recording and registration of the spectra by means of at least one optical device with at least one spectrometer and / or further detectors,

 Generation of digitized signals in the form of data points and storage of the detected spectra in storage units of classification units of an evaluation unit,

 Spectral pretreatment by individually evaluating the stored spectra in the individual storage units and providing the associated digitized evaluated signals for further processing,

 Separation of the pretreated spectra as a training set and as a test set,

- interpretation of the pretreated spectra as a training set and of a test set separated from the training set,

 wherein according to the invention at least one

 - calculation of the classifiers of the series of individual classification methods considered, including iterative procedures and validation in the classification units,

 - Classification of the evaluated spectra of the training set with all classifiers of the series,

Classifying the spectra of the training set into a class of object information with an expression of a class membership probability, Calculation of the median or execution of a cluster analysis for displaying the probability result of the object information belonging to the class of the training set,

 - Classification of the evaluated spectra of the test set with all classifiers of the series,

 Classifying the spectra of the test set into a class of object information with an expression of a class membership probability; and

 Calculation of the median or execution of a cluster analysis for displaying the probability result / classification result of the object information belonging to the class of the test set

 be performed.

In addition, it should be noted that in general the registered spectra 4 are naturally highly variable. This is due, on the one hand, to the inherent variability of biological systems and, on the other hand, to the sensitivity of Raman spectroscopic measurements. External disturbances of a systematic and random nature lead to a high variability of the spectral features and thus superimpose the feature-relevant information. Furthermore, in the method of Raman spectroscopy also fluorescent light occurs, which also contains molecular information, but at the same time superimposes the generally much weaker Raman spectroscopic molecular information on the composition of the examined object 2. On the basis of this preamble and FIG. 1, a further embodiment with more than two features of object information will be explained. 1 is a schematic block diagram of an inventive method 1 for the classification of spectra 4 of an object 2 with complex information content, in particular of optical molecular spectra 4 for the assignment of object information / features 3 for a probable determination of a For example, dual object information 31, 32 or four object information 51, 52, 53, 54 are shown.

 As objects to be examined 2 also tissue samples, such as brain tumors, can be used and used and instead of the dual feature information 31, 32, for example, four different features 3 with 51, 52, 53, 54 selected and determined, for example

> Characteristic 51: healthy tissue,

 > Characteristic 52: Tumor tissue with tumor grade I and II in accordance with

 histological grading scheme of the World Health Organization (WHO),

 > Characteristic 53: tumor tissue with tumor grade III and IV according to WHO and

Feature 54: necrotic tissue.

The recording and registration of the backscatter radiation from the tissue sample is carried out by means of at least one optical device, for example as described in the publication DE 10 2014 010 150 A1. The registered backscatter spectra 4 are digitized and stored in an evaluation unit. The data pretreatment is carried out by, for example, three different methods 5, 6, 7, the data sets obtained may contain, for example, raw spectra, normalized spectra and spectra with a nonlinear baseline correction, wherein the stored spectra are evaluated individually in the individual memory units and the associated digitized evaluated signals for Further processing can be provided. The pretreated spectra are designed as a training set, wherein according to the invention a calculation of the classifiers of the series of the considered individual classification methods is carried out involving iterative methods and a validation in the classification units. The classification of the evaluated spectra of the test set with all classifiers of the series and the classification of the tissue spectra into a class of object information according to the features 51 to 54 with an expression of a probability of Class membership. The classification is evaluated with the calculation of the median or by means of a cluster analysis and the probability result / classification result of the object information belonging to the class of the test set is shown. This means that, for each registered spectrum of a tissue sample, multiple scores are used to calculate a score which, after fixed cut-offs, lies in one of the 4 probability ranges corresponding to the histological findings of the characteristics: 51 - Healthy / 52 - WHO I.II / 53 - WHO III, IV / 54 - correspond to necrosis. A device for classifying spectra 4 of objects 2 having complex information content, preferably with objects in the form of chicken eggs 2 for a determination of dual egg information 31, 32 - female or male - in which the aforesaid method is realized and which largely represents the block (FIG. Box) - representation in Fig. 1 is formed accordingly, may comprise at least the following units

 at least one detecting optical device with at least one spectrometer and / or further detectors for recording and registering the spectra 4,

 a unit for generating digitized signals in the form of data points that realize the spectra 4.

 Storage units for storing the spectra 4 in the classification units / groups 9, 10, 11, 12 of an evaluation unit comprising the classification units,

 - Units for spectral pretreatment 5, 6, 7, 8, in which the stored spectra 4 are evaluated individually in the individual storage units and the associated digitized evaluated signals are provided for further processing,

 - training sets 19 for the design and use of pretreated spectra 4; 25, 26, 27, 28,

at least one classification unit for the groups 9, 10, 11, 12 for the calculation of the classifiers of the series 13, 14, 15, 16 of considered individual conventional classification methods 25, 26, 27, 28 involving iterative methods and validation in the classification units,

 Test sets 29 for classifying the evaluated spectra 4 with all classifiers of the series 13, 14.1 5, 16,

 a unit for classifying the spectra 4 into, for example, a dual class - male 32 or female 31 - of object (egg) information with an expression of a class membership probability,

an evaluation unit for calculating the classification result 18, for example in the form of the median 30 or after carrying out a cluster analysis for representing the probability result of the object (egg) information 31, 32, for example one of the dual class - female or male multiple classification methods are constructed with the four features 51, 52, 53, 54 or with other predetermined features.

LIST OF REFERENCE NUMBERS

1 procedure in a box presentation

 2 object / egg

3 Object Information / Features

 4 registered spectra

 5, 6, 7, 8 pretreatment

 9, 10, 1 1, 12 classification

 13, 14, 15, 15, 16 series of classifiers

131, 132, 133, 134, 135, 136 classifier

 17 rating

 18 classification result

 19 training set

 20 Molecular composition

21 fluorescence intensity

 22 fluorescence profile

 23 variation physical parameters

 24 Classified Test Set

25, 26, 27, 28 pretreated spectra

29 test set with preferably selected 30% of the

 spectra

 30 median

 31 Object information / characteristic female

 32 Object information / characteristic male

33 unshaded face, the female

 Assigned to gender

 34 Pillar of the female sex

 35 Column of the male sex

 36 Hatched face, the male

 Assigned to gender

37 classification result image 38, 39 Column display

 40 weighting chart

 41 Middle data point in the region of one

 spectrum curve

42 Fixed separation limit

 43 first histogram of cluster analysis

 44 second histogram of cluster analysis

 45 = © comparison point

 46 = © comparison point

47 = © comparison point

 48 = ® comparison point

 49 = © comparison point

 50 classifier according to the prior art

51, 52, 53, 54 feature

Claims

claims 1 . A method of classifying spectra of complex information content objects (2) having at least two different object information (31, 32, 51, 52, 53, 54) using a method for registering and pretreating spectral data and a method associated with data pre-treatment the classification with the calculation of a classifier, characterized in that after the registration and the data pretreatment (5, 6, 7, 8) of the spectra a multiple classification method with at least two different methods of data pretreatment (5, 6, 7, 8) of the spectra and the method of classification (9, 10, 11, 12) associated with the respective data pretreatment (5, 6, 7, 8). Method according to Claim 1, in which the following steps are carried out after registration and data pretreatment:  o a calculation of several classifiers of the series (13, 14, 15, 16) per Type of data pretreatment (5, 6, 7, 8),  o a determination of the classifiers of the series (13, 14, 15, 16) iteratively calculated and validated,  o a calculation of probabilities of the class membership, o an equal inclusion of all classifiers of the series (13, 14, 15, 16) or classifiers (131, 132, 133, 134, 135, others for the Series 14, 15, 16) in the determination of a classification result (18, 30, 43, 44). The method of claim 1, wherein in determining the number of calculated classifiers (N _G ) in the series (13, 14, 15, 16) per classification group (9, 10, 1 1, 12) the scope of the spectral data points (vs ) as well as the double half-width of the spectral regions (w _s ) as well as the number of selected spectral regions (R _s ) are taken into account: the equation (I) ensures that each data point can be selected with equal probability. Method according to Claim 3, in which the data points belonging to the circumference of the spectral data points (vs) are weighted. Method according to Claim 1, in which at least one of the spectral pretreatments (5, 6, 7, 8) is structured in such a way that particular features emerge in each case and other specific features are suppressed. Method according to Claim 5, in which at least one spectrum pretreatment (5) with identically determined and equally weighted features is added to at least one of the spectral pretreatments (6, 7, 8) with differently determined characteristics of the evaluation. Method according to Claim 1, in which the pretreated spectra (4) are designed as a training set (19) and a plurality of classifiers of the series (13, 14, 15, 16) or classifiers (131, 132, 133, 134, 135, etc.) are determined and be validated. The method of claim 1, wherein at least one method of unsupported classification and / or supported classification for selecting spectral regions R _s or individual wavelength ranges and subsequent analysis are used. Method according to claim 1, wherein as a method for classification in the classification groups (9, 10, 11, 12) a method of the neural networks and / or a method of linear time-frequency transforms (English: Wavelets) are used. 10. The method of claim 1, wherein pretreated spectra (5, 6, 7, 8) are classified by optical molecular spectroscopy, preferably absorption, emission, scattering or UV / vis, NIR, IR absorption, fluorescence, Raman. 1 1. Method according to Claim 1, in which raw spectra (25), baseline corrections (26), normalizations (27), derivatives, covariance and / or Raman spectra (28) are used for data pretreatment (5, 6, 7, 8). 12. The method of claim 1, wherein for the evaluation of the classifiers of the series (13, 14, 15, 16) for a classification result (18), a calculation of a median (30) or a cluster analysis are performed. 13. The method according to claims 1 to 12, wherein the following steps are carried out:  - Recording and registration of the spectra (4) by means of at least one optical device with at least one spectrometer and / or further detectors,  Generation of digitized signals and storage of the registered spectra (4) in memory units of the classification units / groups (9, 10, 11, 12) of an evaluation unit,  Spectral pretreatment (5, 6, 7, 8) by individually pretreating the registered and stored spectra (4) in the individual storage units and providing the associated digitized evaluated signals for further processing, Separation of the pretreated spectra as training set 19 and as test set 29, Design and use of the pretreated spectra as training set 19 and of a test set 29 separate from the training set 19,  - Calculation of the classifiers of the series (13, 14, 15, 16, 131, 132, 133, 134, 135) of the individual classification methods considered, including iterative methods and validation in the classification groups (9, 10, 11, 12 )  Classification of the pretreated spectra of the training set with all classifiers of the series,  Classifying the spectra of the training set into a class of object information with an expression of a class membership probability,  Calculation of a classification result by calculation of the median or by performing a cluster analysis for representing the probability result of the object information belonging to the class of the training set,  Classification of the pretreated spectra of the test sets (24) with all classifiers of the series (13, 14, 15, 16),  Classifying the spectra into a class of object information (31, 32) with an expression of a class membership probability, - Calculation of a classification result (18, 30, 43, 44) by calculation of the median (30) or by performing a cluster analysis (43, 44) for displaying the probability result of the object information associated with a class (31, 32). 14. Method according to claims 1 to 13, in which bird eggs, preferably chicken eggs, are used as objects (2) and the dual information about the female gender (31) and about the male gender (32) is used as object information. 15. The method according to claims 1 to 13, wherein as objects (2) optionally tissue samples, such as brain tumors, are used and as object information four features (3) with:  Feature (51): healthy tissue,  Feature (52): tumor tissue with tumor grade I and II corresponding to  histological grading scheme of the World Health Organization (WHO),  Feature (53): tumor tissue with tumor grade III and IV according to WHO and  Feature (54): necrotic tissue  used and selected and determined. 16. The method of claim 1, wherein after a passage of at least two each with a previous spectrum different data pretreatment (5, 6, 7, 8) provided classification method (9, 10, 1 1, 12) with at least one determined classifier (131, 141 , 151, 161) collectively at least two determined classifiers (131, 141, 151, 161) for the evaluation (17) and the subsequent determination of a probability result (18) with respect to the predetermined different object information (31, 32; 51, 52, 53, 54 ) and used, wherein the probability result (18) is output, so that an inference is made possible at least on the object information (31, 32; 51, 52, 53, 54) determined with a highest value. 17. A device for classifying spectra (4) of objects (2) with complex information content, in which the aforementioned method according to one of claims 1 to 16 is realized, comprising at least the following units at least one detecting optical device with at least one spectrometer and / or further detectors for recording and registering the spectra (4), - A unit for generating digitized signals in the form of data points that realize the spectra (4).  Storage units for storing the registered spectra (4) in the classification units / groups (9, 10, 11, 12) of an evaluation unit comprising the classification units,  - Units for spectral pretreatment (5, 6, 7, 8) in which the registered spectra (4) are pretreated individually in the individual storage units and the associated digitized evaluated signals are provided for further processing, Training sets (19) for designing and using the pretreated spectra (- 25, 26, 27, 28),  At least one classification unit for the groups (9, 10, 11, 12) for the calculation of the classifiers of the series (13, 14, 15, 16) including iterative methods and a validation in the classification units,  Test sets (29) for classifying the pretreated spectra (25, 26, 27, 28) with all classifiers of the series (13, 14, 15, 16),  a unit for classifying the pretreated spectra (25, 26, 27, 28) into at least one dual class of object information (31, 32, 51, 52, 53, 54) with an expression of a class membership probability, a rating unit for calculating the classification result (18, 30, 43, 44), e.g. in the form of the median (30) or performing a cluster analysis (43, 44) for determining the probability result of at least one of the predetermined object information (31, 32, 51, 52, 53, 54).